pocketflow/README.md

3.5 KiB
Raw Blame History

Pocket Flow - LLM Framework in 100 Lines

License: MIT Docs


A 100-line minimalist LLM framework for (Multi-)Agents, Workflow, RAG, etc.

Why Pocket Flow?

Pocket Flow is part of a New Development Paradigm: Let LLM Agents (e.g., Cursor AI) build LLM Agents for humans!

Now, if you use Cursor AI...

  • 🫠 with frameworks like LangChain: You will run into errors (e.g., using deprecated packages, hallucinating new functions, or failing to express new agentic designs).

  • 😐 without any framework: Ironically, this often works better. However, the code it produces is ad hoc and NOT modular or maintainable for large projects.

  • 🥰 with Pocket Flow: Pocket Flow is minimal and expressive, so it initially works similarly to the "no framework" approach. But as a project grows, Pocket Flow:

    • Nodes and flows enforce a modular design.
    • A shared store enforces the separation of data and compute logic.

So, the 100-line limit compels Cursor AI to follow good design principles without sacrificing expressiveness, which is essential for real-world projects.


To set up:

What Is Pocket Flow?

The 100 lines capture what we believe to be the core abstraction of LLM projects:

  • Computation: A graph that breaks down tasks into nodes, with branching, looping, and nesting.
  • Communication: A shared store that all nodes can read and write to.


From there, its easy to implement popular design patterns like (Multi-)Agents, Workflow, RAG, etc.



  • To learn more about how it works, check out the documentation
  • For an in-depth dive into the design, check out the essay